Scholarlink

<p>Segmentation of spinal motion segments from multi‑slice computed tomography (MSCT) images is essential for clinical evaluation and pre‑operative planning. This study presents a fully automated framework that combines a three‑dimensional U‑Net for MSCT segmentation, a watershed‑based algorithm for vertebra‑wise separation, and a marching cubes pipeline for three‑dimensional reconstruction. A dataset of anonymized lumbar‑spine MSCT scans from 50 male patients was used for model development. The proposed 3D U‑Net achieved a Dice similarity coefficient of 0.96, a Jaccard Index of 0.93, a Hausdorff distance of 3.1&#xa0;mm, a precision of 0.97, and a recall of 0.95, outperforming Dense‑U‑Net and nnU‑Net baselines trained on the same data. Vertebra‑separation performance reached a 95.8% success rate with a 3.4% mis‑segmentation rate and a mean boundary deviation of 1.9&#xa0;mm. An ablation study confirmed that posterior‑arch removal, ROI detection, watershed separation, and data augmentation each contributed substantially to overall accuracy. A graphical user interface was developed to enable interactive inspection, manipulation, and export of patient‑specific spinal motion‑segment models. The system provides a reliable and clinically practical solution for MSCT‑based segmentation and 3D visualization, supporting improved surgical planning and potential integration into orthopedic workflow pipelines. Future work will extend the framework to multi‑center datasets and dynamic (4D) motion analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Segmentation and 3D Visualization of Spinal Motion Segments from MSCT Images Using a 3D U‑Net Framework

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Antor Mahamudul Hashan,
  • Khlebnikov Nikolai Alexandrovich,
  • Denis Protasov

摘要

Segmentation of spinal motion segments from multi‑slice computed tomography (MSCT) images is essential for clinical evaluation and pre‑operative planning. This study presents a fully automated framework that combines a three‑dimensional U‑Net for MSCT segmentation, a watershed‑based algorithm for vertebra‑wise separation, and a marching cubes pipeline for three‑dimensional reconstruction. A dataset of anonymized lumbar‑spine MSCT scans from 50 male patients was used for model development. The proposed 3D U‑Net achieved a Dice similarity coefficient of 0.96, a Jaccard Index of 0.93, a Hausdorff distance of 3.1 mm, a precision of 0.97, and a recall of 0.95, outperforming Dense‑U‑Net and nnU‑Net baselines trained on the same data. Vertebra‑separation performance reached a 95.8% success rate with a 3.4% mis‑segmentation rate and a mean boundary deviation of 1.9 mm. An ablation study confirmed that posterior‑arch removal, ROI detection, watershed separation, and data augmentation each contributed substantially to overall accuracy. A graphical user interface was developed to enable interactive inspection, manipulation, and export of patient‑specific spinal motion‑segment models. The system provides a reliable and clinically practical solution for MSCT‑based segmentation and 3D visualization, supporting improved surgical planning and potential integration into orthopedic workflow pipelines. Future work will extend the framework to multi‑center datasets and dynamic (4D) motion analysis.